DeepSeek once operated with the quiet confidence of a lab born from a hedge fund. Its models delivered strong performance at low cost. Chinese developers praised them. The startup largely avoided the spotlight and heavy outside capital.
That changed in April. A preview of Anthropic’s latest model, known as Mythos, reached executives at the Hangzhou-based lab. CEO Liang Wenfeng saw capabilities that forced a reassessment. He concluded his team needed far more resources to stay competitive. The result? DeepSeek’s first major outside funding round. It closed at roughly $7.4 billion.
The funding marks a sharp pivot for one of China’s most efficient AI developers.
Details emerged first in reporting by The Information. The preview of Mythos convinced Liang that without a much larger cash pile, DeepSeek could not match the frontier pace set by top U.S. labs. Plans now include doubling the workforce and expanding reliance on domestic Huawei chips to bypass U.S. export limits.
The sum ranks among the largest single venture rounds in Chinese tech history. Bloomberg reported in early June that DeepSeek was nearing a deal of about 50 billion yuan, or $7.4 billion, with Tencent Holdings and Contemporary Amperex Technology Co. (CATL) as the biggest backers. The state-linked National Artificial Intelligence Industry Investment Fund also joined. (Bloomberg, June 3, 2026)
But the final structure stands out even more. According to follow-on coverage, Liang routed much of the capital through a limited partnership he controls. Most investors receive no voting rights. A five-year lock-up applies. One investor reportedly holds a single exception. The post-money valuation exceeded $50 billion. That represents a jump from earlier estimates around $10 billion.
Liang, who founded DeepSeek after working in quantitative trading, had long preferred self-funding and internal resources. The shift came suddenly. One preview. A demonstration of reasoning depth, coding fluency and efficiency that outstripped expectations. DeepSeek’s own models had gained attention for outperforming larger rivals on certain benchmarks while using fewer resources. Yet the gap appeared wider than anticipated.
And the pressure isn’t abstract. U.S. labs benefit from access to the latest Nvidia hardware. Chinese teams face export controls. DeepSeek’s answer involves scaling up with Huawei’s Ascend processors. The new capital will fund massive compute clusters built around those chips. It will also support hiring drives across research, engineering and infrastructure teams. Doubling headcount from current levels signals ambition to rival the thousands-strong organizations at Anthropic, OpenAI and Google DeepMind.
Recent coverage highlights how this round reflects broader tensions. The Information detailed the unusual governance terms that keep power concentrated with Liang. Investors accepted the arrangement because of DeepSeek’s track record and the strategic weight of advanced AI in Beijing’s priorities. Tencent, CATL, NetEase, JD.com, IDG Capital, Monolith Capital and the national AI fund all participated, according to investor trackers and secondary reports.
Short sentences tell part of the story. Massive capital. Domestic chips. Founder control. These elements define the new phase.
Yet questions linger. Can DeepSeek translate fresh billions into models that close the perceived gap with Mythos? Early signs suggest the lab already trains on hybrid infrastructure. Its prior releases demonstrated clever optimization. Now the bet is that more data, more compute and more talent will produce frontier results without full access to restricted U.S. technology.
The timing also matters. Anthropic itself raised enormous sums earlier in 2026, reaching a valuation above $900 billion in one round before pursuing an IPO. Its Mythos release, initially limited to trusted U.S. partners and organizations, carried both technical and geopolitical weight. Reports indicate the model’s preview circulated among select audiences, prompting reactions far beyond American borders.
DeepSeek is hardly alone. Other Chinese labs such as Moonshot AI and MiniMax have drawn scrutiny from U.S. companies over alleged model distillation and account misuse. Anthropic has publicly accused several Chinese entities of creating thousands of fraudulent accounts to extract knowledge from Claude models. Those disputes add friction to an already fragmented global race.
But the DeepSeek move stands apart. It wasn’t regulatory pressure or a leaked paper that triggered action. It was a direct view of a competitor’s progress. Liang’s decision to raise outside money for the first time signals a belief that the margin for error has narrowed. Efficiency alone may no longer suffice.
So what comes next? DeepSeek intends to accelerate model development on multiple fronts. Larger context windows. Stronger agentic capabilities. Improved multilingual performance. All while navigating hardware constraints that U.S. labs largely avoid. The Huawei pivot forms a core part of that strategy. Success here could validate China’s push for semiconductor self-reliance. Failure would underscore the cost of restricted access to leading chips.
Industry observers note the valuation jump. From roughly $10 billion earlier this year to more than $50 billion post-money. That compression of timelines reflects how quickly perceptions shift when a single demonstration lands. Similar dynamics played out when earlier Chinese models like DeepSeek-V2 or Qwen variants posted impressive benchmark scores. Each time, Western labs responded with bigger training runs.
This episode reverses the gaze. A U.S. breakthrough, even in preview form, sent shock waves through a leading Chinese lab. The response arrived in the form of one of the biggest checks ever written for an AI startup in China.
Analysts following the deal point to the unusual terms as a sign of founder confidence. Liang apparently secured commitments while retaining decisive control. That structure may become a template for other Chinese AI firms wary of foreign influence or board-level dilution. Yet it also places enormous responsibility on one executive and his team to deliver results that justify the optimism.
Recent discussions on X, including posts from The Information’s own account, underscore the surprise factor. “Anthropic’s Mythos preview convinced CEO Liang Wenfeng that China’s top AI lab needed far more money to compete,” the publication noted in multiple updates over the past week. The story continues to generate debate among developers and investors tracking frontier model progress.
Beyond the numbers sits a larger contest. Talent. Energy. Data. Compute. Each element now carries strategic importance. DeepSeek’s expansion will test whether concentrated domestic capital and engineering ingenuity can overcome hardware asymmetries. Its success or shortfall will influence how governments and companies allocate resources in the years ahead.
The lab’s hedge-fund origins once shaped a culture of measured bets and high efficiency. That mindset helped produce capable models on modest budgets. Now the budget is no longer modest. The bets have grown dramatically larger. The question is whether the culture adapts fast enough to match the scale.
Executives at U.S. labs have watched the development closely. Some privately acknowledge that Chinese teams have repeatedly surprised them with rapid iteration. Others see the $7.4 billion infusion as validation that the competitive threat remains real. Either way, the pace shows no sign of slowing.
DeepSeek has not issued a formal statement on the funding’s connection to Mythos. Public comments from Liang remain limited. Yet the sequence of events, as reported, leaves little doubt. One preview. One realization. One record-breaking capital infusion. The AI race grows only more intense.
WebProNews is an iEntry Publication